Research On Waveform Classification Method Based On 3D Seismic Data

With effective waveform classification technology, we can seek out the correspondence between seismic facies and the geological structure, estimate rock properties and changes of rocks in lithology, perform regional stratigraphic interpretation, obtain the depositional system, and predict the reservoir facies belts and the area where is beneficial for oil. Thereby, we are clear about the late mining goals, which can improve the efficiency of exploration and exploitation, reduce the cost, and enhance the ability of stable energy supply.This thesis described the background and practical significance of the volume waveform classification, deeply analized the related technologies and research status, and concluded the strengths and weaknesses of these technologies. We studied the feature dimensionality reduction, clustering algorithms, similarity measurement between seismic trace waveforms and the updating of a cluster's center. We proposed two different three-dimensional volume waveform classification algorithms. The main works of this thesis are as follows:Firstly, original seismic data contains redundant information which interferes the classification. Traditional dimension reduction methods cannot keep the information in the non-linear structure. This paper proposed a volume waveform classification method based on intrinsic feature analysis. In this method, seismic trace is treated as a point of high-dimensional space, and the relationship between adjacent points is characterized by Laplace matrix. Based on these relationships and aimed at minimizing the smoothness of the manifold, we solve a generalized eigenvalue problem, and calculate a set of nonlinear basis functions with the corresponding eigenvectors. With the nonlinear basis functions, points in high-dimensional space can be mapped into low dimensional space and we can get the low dimensional manifolds embedded in the ambient observation space. Then SOM is used to get the final classification result. Intrinsic Feature Analysis well preserved the intrinsic features of original seismic data, helped to distinguish seismic traces. SOM enhanced the ability to identify different waveform categories, and more realistically reflected the underground structures.Secondly, as the commonly used method to measure the similarity between samples can only deal with sequences of same size. When the horizon is not accurate and not aligned, this method is not able to characterize the degree of similarity between samples accurately. To solve this problem, combining with the fact that original seismic traces are time series, this paper proposed a volume waveform classification based on Dynamic Time Wrapping. Dynamic Time Wrapping is introduced to the volume waveform classification as the measurement of similarity between seismic traces. With the combination of the most basic clustering algorithm--partition and a global averaging method for dynamic time warping, a new volume seismic waveform classification is formed. This method overcomes the limitations of traditional similarity measurement methods, can describe the similarity more accurately, overcomes the effects of the errors caused horizon interpretation, and improves the reliability of the result.Finally, the above mentioned algorithms are experimentally verified in many three-dimensional seismic datas of actual areas. And we compared the results with those of the traditional methods. It is proved that the methods proposed in this paper are more efficient, which can truly reflect the changes and the distribution of underground structures.